Federal learning method using synonymous data

The invention provides a federal learning method using synonymous data, comprising: a coordination device sending a general model to each client device, each client device executing a training program, comprising: encoding private data into an abstract by an encoder, training a client model according to the private data, the abstract and the general model, and sending the abstract and customer parameters of the customer model to a coordination device, the coordination device judging an absent customer device in the customer devices, generating synonymous data by a synonymous data generator according to the abstract corresponding to the absent customer device, and training a substitution model by the coordination device according to the synonymous data and the abstract corresponding to the absent customer device, and the coordination device executes aggregation operation according to the replacement model parameters of the replacement model and the client parameters of each client device except the absent clie